Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Bài kiểm tra chỉ số ổn định dân số (PSI)

Tính toán PSI từ các lượt chia sẻ đã được gộp và kiểm tra xem lựa chọn tham chiếu, làm mịn số 0 và việc gộp ảnh hưởng đến việc diễn giải như thế nào.

Đường dẫn hướng dẫn liên quanChỉ số ổn định dân số Psi
Câu hỏi 1 của 8

What quantities enter a common PSI formula for each bin?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. What quantities enter a common PSI formula for each bin?
    • The share difference multiplied by the log ratio of comparison to reference shares
    • The raw count difference squared
    • The model's true-positive rate and recall
    • The number of bins times mean prediction error
  2. What happens if one reference bin share is zero in the PSI formula?
    • The PSI is automatically zero
    • The log ratio is undefined and requires a stated smoothing convention
    • The bin can be ignored without changing the estimate
    • The comparison share becomes the model accuracy
  3. Why keep bin boundaries fixed across reference and comparison periods?
    • It makes PSI independent of sample size
    • It guarantees each period has equal bin counts
    • It makes changes in proportions comparable across the same ranges
    • It converts PSI into a hypothesis test
  4. What can coarse binning do to a real distribution change?
    • Turn a feature into a label
    • Guarantee a larger PSI
    • Remove the reference sample
    • Hide within-bin differences and produce a small PSI
  5. What does a high PSI establish by itself?
    • Model accuracy has definitely declined
    • Binned distributions differ under the chosen setup
    • The shift was caused by a specific feature
    • The model is unfair
  6. Why are common PSI thresholds not universal guarantees?
    • Thresholds change the underlying feature values
    • PSI is always negative
    • Interpretation depends on binning, sample and application context
    • Every domain uses identical reference populations
  7. Why inspect per-bin PSI contributions?
    • They prove which population caused the change
    • They provide exact model accuracy for each bin
    • They remove the need for labels
    • They show which ranges account for the overall shift
  8. Which signal can change without PSI changing much?
    • A conditional relationship between features and outcome
    • The fixed reference bin boundaries
    • The comparison sample proportions by definition
    • The PSI formula itself